Disguising a Reference So Others Can Test Your Filter

[Bret Cahill]

Say someone boasts about having a new filter that, in certain
situations, can reduce noise more and faster than any other filter?

Is there any way to encode a reference that isn't directly
identifiable but still useful as a reference? That way the reference
could be posted along with the noise for the test.

One way might be to make the noise in the reference the negative of
the noise in the signal times some factor. Without the correct
factor, without the SNR, it could be difficult or impossible to
recover the original reference or signal.


Bret Cahill

 

[whit3rd]

On Monday, March 14, 2011 4:09:19 PM UTC-7, Bret Cahill wrote:

Say someone boasts about having a new filter that, in certain
situations, can reduce noise more and faster than any other filter?

But 'reduce noise' isn't something a filter does. It separates a
known type of signal from a signal + distortion + noise input,
and ONLY if the signal is known in character can the
filter function be evaluated.

Reducing noise/signal ratio usually results in the as-filtered
data having less degrees of freedom than the initial data, i.e.
it reduces a hundred points (one hundred nearly-independent
numbers) to three numbers. There's potentially only
3 noise components in the result, so the noise is lessened
(by a factor of perhaps sqrt(100/3), while the signal can
be nearly complete (down by a factor of circa 1.0).

 

[dbd]

On Mar 14, 4:09 pm, Bret Cahill <Bret_E_Cah...@yahoo.com> wrote:

Say someone boasts about having a new filter that, in certain
situations, can reduce noise more and faster than any other filter?

Is there any way to encode a reference that isn't directly
identifiable but still useful as a reference?  That way the reference
could be posted along with the noise for the test.

One way might be to make the noise in the reference the negative of
the noise in the signal times some factor.  Without the correct
factor, without the SNR, it could be difficult or impossible to
recover the original reference or signal.

Bret Cahill

The formulation of your question is a little less than complete and
coherent. But you might get an answer by clarifying your terms and
purpose.

What do you mean by "reference"?

How does the "reference" relate to testing a filter?

What parameters do you wish to demonstrate?

What information are you trying to obscure?

Dale B. Dalrymple

 

[glen herrmannsfeldt]

In comp.dsp Bret Cahill <Bret_E_Cahill@yahoo.com> wrote:

Say someone boasts about having a new filter that, in certain
situations, can reduce noise more and faster than any other filter?

Is there any way to encode a reference that isn't directly
identifiable but still useful as a reference? That way the reference
could be posted along with the noise for the test.

(snip)

As I replied to Jerry's post, see the wikipedia page Trusted_client.

The problem of distributing software source, such that the
recipient can use it but not modify or reverse engineer it
comes up often, but there is no easy answer. (More often
for HDL (hardware description language) code, but it is
the same problem.

-- glen

 

[Rune Allnor]

On Mar 15, 12:09 am, Bret Cahill <Bret_E_Cah...@yahoo.com> wrote:

Say someone boasts about having a new filter that, in certain
situations, can reduce noise more and faster than any other filter?

Any one in particular you have in mind?

Is there any way to encode a reference that isn't directly
identifiable but still useful as a reference?  That way the reference
could be posted along with the noise for the test.

The usual way to do these things is to publish a method
of design in one of the journals. That way the algorithm
to design the filter is reviewed by people who ought to
know the subject at hand.

One way might be to make the noise in the reference the negative of
the noise in the signal times some factor.  Without the correct
factor, without the SNR, it could be difficult or impossible to
recover the original reference or signal.

What use would the filter be, then? It doesn't work unless
you know the exact noise that corrupts the signal? It's a
ridiculous idea.

Rune

 

[aruzinsky]

On Mar 14, 5:09 pm, Bret Cahill <Bret_E_Cah...@yahoo.com> wrote:

Say someone boasts about having a new filter that, in certain
situations, can reduce noise more and faster than any other filter?

That's not a question. Is English your native language?

Is there any way to encode a reference that isn't directly
identifiable but still useful as a reference?  That way the reference
could be posted along with the noise for the test.

But, why would you want it to not be directly identifiable?

One way might be to make the noise in the reference the negative of
the noise in the signal times some factor.

Is this a one dimensional signal or an image? I bet that I can detect
such an image "reference" noise by visual inspection with the "signal"
noise.

Without the correct
factor, without the SNR, it could be difficult or impossible to
recover the original reference or signal.

Wrong.

Y1 = X + E

Y2 = C*E

where C is unknown scalar, X = uncorrupted signal matrix, E = noise
matrix

Solve:

min || Y1 + A*Y2 ||
A

A = -1/C

E = -A*Y2

X = Y1 + A*Y2

 

[Rich Grise]

Bret Cahill wrote:

Say someone boasts about having a new filter that, in certain
situations, can reduce noise more and faster than any other filter?

Is there any way to encode a reference that isn't directly
identifiable but still useful as a reference? That way the reference
could be posted along with the noise for the test.

One way might be to make the noise in the reference the negative of
the noise in the signal times some factor. Without the correct
factor, without the SNR, it could be difficult or impossible to
recover the original reference or signal.

Just get any ol' sound source, like a recording of "The 1812 Overture,"

and sum it with some pseudorandom noise, such that either could be
recovered with a correlator of some kind?

Good Luck!
Rich

 

[Rich Grise]

Rune Allnor wrote:

What use would the filter be, then? It doesn't work unless
you know the exact noise that corrupts the signal? It's a
ridiculous idea.

Brett has a lot of these.

Hey, at least he's thinking!

Cheers!
Rich